@ghopper hard to say where the problem is

from reading the backlog, way too many variables consider

@r0man I really don’t have any good ideas about that one - so I don’t really have any guidance to offer - it’s why I punted on it

@dnolen I'll try creating a minimal example later. 👍:skin-tone-2:

@dnolen the ideas I have is to recompile/load the loader when load-file or evaluate-form are called. here my problem is that I don't know how to get all inputs. The other one is, to only compile the loader on a full compile. do you remebere this one: https://github.com/clojure/clojurescript/commit/427aaba6f557df90e0ad3c25b15531d353fd0ca3

there was a check in there to not compile the loader. I wonder if this made things even worse when having a repl attached to a project. So I was thinking about not recompiling loader on a load-file/evaluate-form, to not wipe out the module info, and only compile the loader on a full compile.

@r0man thinking about it some more I don’t think we need to bother with recompiling

we just need to provide updated modules graph

but the tricky part is getting all inputs - which is still very messy

so perhaps the right answer for now is to just avoid compiling the loader

and when would we avoid it? compile it initially after all sources have been compiled, and then skip it afterwards? how could I detect this, when the loader is in the output directory?

I think we just shouldn’t recompile it via the REPL for now

that’s the problem - at the REPL you won’t get all inputs

that’s why it doesn’t work

ok, I'll try look into this.

actually

hrm, an alternative would be to make sure that cljs.loader is always merging

then it will work at the REPL

@r0man ^

you mean merging module-info and module-urls?

yes

so instead of just replacing the constants - it always merges with the previous value

if it exists

ok, I'll try this out as well

I think that should work